Youth and covid-19

impacts on jobs, education, rights and mental well-being

Technical annex _______________________________________________________________

Global Survey on Youth and Covid-19

Technical Annex The Global Survey on Youth and COVID-19 (the Survey), was conducted by partners of the Global Initiative on Decent Job, including the International Labour Organization, the United Nations Major Group for Children and Youth, AIESEC, the European Youth Forum, the European Union Emergency Trust Fund for Africa and the Office of the United Nations High Commissioner for Human Rights. The online survey, available in 23 languages, was conducted between 21 April and 21 May 2020 and taken by 12,605 individuals aged 18–34 years from 112 countries.1

This technical annex describes the survey promotion in Section I, the dataset and key outcome variables in Section II, explains the survey weighting methodology and presents alternative approaches in Section III, examines the representativeness of the survey sample with regard to survey respondents’ educational background and income in Section IV, and presents the results of robustness checks in Section V.

I. Survey Promotion

The Survey was translated and available in 23 languages including Arabic, Bengali, Chinese, Filipino, French, German, Hausa, Hindi, Indonesian, Italian, Japanese, Korean, Polish, Portuguese, Romanian, Russian, Sinhala, Spanish, Swahili, Thai, Turkish and Vietnamese. Translations were done by volunteers and each language version was pilot tested. The survey platform SurveyMonkey was used to conduct the Survey. Dissemination and promotion of the Survey was first done through social media and then through direct emailing of young people who were identified to be amongst the networks of the five survey partners.

II. The Dataset

The micro-survey data resulting from individual responses are merged with additional macro-level information on (i) ILO regions (Africa, Americas, Arab States, Asia and the Pacific, Europe and Central Asia)2, (ii) World Bank income groups (low, lower-middle, upper-middle, high income countries)3 as well as (iii) UN Operational rates of exchange4 to convert self-declared gross monthly income (provided in the local currency) to US Dollar.

As shown in Table A-1 the survey included modules on employment, education and training, mental well-being, rights, social activism, policy responses and personal information. In total, the survey contained 44 questions. Survey respondents completed the survey on 1 Only respondents at least 18 years old were included in the final dataset as for youth below the age of 18 parental consent could not be assured through the online survey. Moreover, for 416 respondents weights could not be computed based on the country, gender and age combination of the respondent (Section II for details on how weight construction). Moreover, only observations from countries with at least 10 respondents were retained (affecting 211 observations). 2 Regional breakdown of ILO Member States is available at: https://www.ilo.org/global/regions/lang--en/index.htm 3 The most recent 2020 dataset was used and is available at: https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups 4 Rates valid on 1 May 2020 were used and are available at: https://treasury.un.org/operationalrates/OperationalRates.php

average in 10 minutes and the completion rate was 65 per cent. Survey respondents are asked to indicate their “current situation” with a choice between (i) “working”, (ii) “studying”, (iii) “working and studying”, (iv) “not working or studying”. Depending on their current situation, survey respondents completed either the employment (1a) and/or education and training (1b) or the ‘not working and job loss’ (1c) module. All respondents were then directed to complete modules 2 to 6.

Description of selected outcome variables:

- Stopped working: This is a dummy variable that takes the value of one for all respondents who (i) declared to be “not working or studying” AND further affirmed to have lost their previous job since the start of the coronavirus outbreak; (ii) declared to be working5 AND further declared to be working at least one hour on a typical day before the outbreak AND reported to be working zero hours on a typical day during the outbreak. The dummy takes the value of zero for all respondents who declared to be working at least one hour on a typical day during the outbreak.

- Hours worked: respondents who indicated to be working were asked to numerically input both their daily hours worked before the outbreak started and during the outbreak.

- Occupation: Survey participants who indicated to be working or who previously worked were asked about their occupation (open-ended question). Of the 4,818 answers that were received,6 3,646 could be manually matched to the International Standard Classification of Occupations (ISCO-08)7 at the two-digit level.8

- Mental wellbeing: the survey featured a module with the Short Warwick Edinburgh Mental Wellbeing Scale (SWEMWBS).9 The SWEMWBS consists of seven items that each contain one positively worded statement, with five response categories from ‘none of the time’ to ‘all of the time’. Answers are scored and summed up, resulting in an index with scores from 7 to 35 with higher scores indicating better mental wellbeing. The raw index is converted into a metric score following guidance from the Warwick Medical School.10

-

Table A-1: Overview of the survey structure Module Taken by Introduction

(country of residence, gender, age, main activity)

All survey respondents

1a Employment Current situation: working OR working & studying

1b Education and training Current situation: studying OR working & studying

5 Or “working and studying”. 6 Out of 6,259 respondents who were asked either about their current or most recent occupation. 7 https://www.ilo.org/public/english/bureau/stat/isco/isco08/ 8 The analysis presented in the main report is limited to a breakdown at the one-digit level. 9 “Short Warwick Edinburgh Mental Wellbeing Scale (SWEMWBS) © NHS Health Scotland, University of Warwick and University of Edinburgh, 2008, all rights reserved.” 10 The conversion table is available at: https://warwick.ac.uk/fac/sci/med/research/platform/wemwbs/using/howto/swemwbs_raw_score_to_metric_score_conversion_table.pdf

1c Not working and job loss Current situation: neither working nor studying 2 Mental wellbeing All survey respondents 3 Social activism All survey respondents 4 Rights All survey respondents 5 Policy responses All survey respondents 6 Personal information All survey respondents

III. Weighting methods

While surveys seek to recruit representative samples of the underlying population, it is common that in the process of sampling, biases arise. According to Fotini, Evangelia and Michail (2013)11, three potential sampling biases might be present in any survey: (i) bias from the under-/over-representation of some subgroups of the population, (ii) non-response bias when these are not random, and (iii) population size bias when comparing two or more countries. Biases can be corrected through the following weighting systems:12

- Non-response weighting: in voluntary surveys, such as the Global Survey on Youth and COVID-19, one of the major threats for the accuracy of the estimates is non-responsiveness by surveyed individuals. This leads to bias if the non-respondents are a non-random sample of the underlying population. In this case, non-response weighting is used to compensate for the fact that persons with certain (known) characteristics are not as likely to respond to the survey.13 Choosing which variables to use in the weighting process is crucial to correct successfully for the biased sample.14

- Population size weighting is used when examining a combination of survey data from two or more countries. It assigns the surveyed sample from each country a weight proportional to the population of the respective country. Population weights are the same for all survey participants within a country, but differ across countries.15

The weighting methodology adopted in the analysis of the sample of the Survey combines elements of non-response weighting and population size weighting. To correct for non-response bias, we weight the population by sex (women, men) and age groups (18-29, 30-34). To correct for population biases we allow weights to be a function of the youth population of each country.

Weights are constructed in three steps:

1) Identification of the relevant population: The relevant population is individuals between the age of 18 and 34 living in the 112 countries represented in the survey.

11 Fotini, T., Evangelia, V., & Michail, V. (2013): Weighting of responses in the Consumer Survey : Alternative approaches – Effects on variance and tracking performance of the Consumer Confidence Indicator Table of Contents. Foundation for Economic & Industrial Research. 12 In addition, design weighting can be used to correct for under-/over- representation of certain subgroups in the sample compared to the population. This is typically adapted when some sub-groups are deliberatelry over/under-sampled, a case that does not apply here. 13 Fotini, Evangelia and Michail (2013), see footnote 9. 14 Pew Research Center (2018): For Weighting Online Opt-In Samples, What Matters Most? 15 European Social Survey. (2014): Weighting European Social Survey Data. Available at: https://www.europeansocialsurvey.org/docs/methodology/ESS_weighting_data_1.pdf

In the final, unweighted sample, 54 per cent of those 18 to 29 old have attained tertiary education (see Table A-5). Of those young people without tertiary education, 85 per cent are studying. Within the 30-34 age cohort, 89 per cent of respondents report having attained tertiary education.16 Given the survey is overwhelmingly representing individuals who either already attained a tertiary degree or are currently pursuing one, we narrow down the sub-population to those youth who attained an advanced degree. It is important to note that not all young people in our sample have already attained tertiary level education, yet this sub-population appears as the best available proxy for the youth population captured by the sample.17

2) Obtaining population weights from the ILO Department of Statistics (ILOSTAT): As the data source, we use the ILOSTAT dataset on (youth) working-age population by sex, age and education.18 We construct population weights for four population bins, dividing the survey respondents by age group (18-29; 30-34) and sex (women, men). We use ILOSTAT datasets on (youth) working-age population by sex and age, as well as (youth) working-age population by sex, age and education.

- For the 18-29 age bin (by sex) we use data on youth working-age population for those with an advanced degree. ILOSTAT only provides data for the 15-29 cohort which is used as an approximation.19 For 28 countries the 15-29 cohort is not available. In these cases we use the two 10 year cohorts 15-24 and 30-34 and take the sum of the 15-24 cohort plus 50 per cent of the 25-34 cohort as an approximation.20

- For the 30-34 age bin (by sex) we use data on working-age population for those with an advanced degree. ILOSTAT provides data for the 25-34 cohort as well as the 25-29 age cohort. We subtract the two to obtain an estimate of the 30-34 age bracket.

- For 18 countries, data on working-age population (by 5 and 10 year age brackets) is available but no breakdown by level of education. We impute population data by modelling the share of the population with a tertiary degree for the 15-29 age cohort and the 30-34 cohort for women and men as an average of all countries with available data in the same ILO region and same World Bank country income group.

- The above procedure allows us to calculate population weights for two age-cohorts (18-29, 30-34) and by sex (women, men) for 112 countries covering 98.8 per cent of all survey respondents.21

3) Calculation of weights: three sets of weights are constructed

16 Among the 11 per cent of respondents in the 30-34 year old age group a significantly lower share, 29 percent, are still studying (or studying while also working). 17 Because we do not use educational background as a weighting variable (also due to a sizable non-response rate), the main reason to condition youth population to those with advanced education is to avoid imbalances where (well educated) respondents from countries with large youth population attain an outsize weight. 18 Available at: https://ilostat.ilo.org/ 19 This is unproblematic: the number of young people 15 to 17 holding an advanced degree is presumably negligible. Furthermore, we are not computing weights for sub-age groups within the 18-29 groups. 20 This assumes that the population is uniformly distributed over the 10 years of the bin. Under the same assumption, the age bin 30-34 can be calculated as the remaining 50% of the 25-34 bin 21 152 observations need to be dropped because no weights could be calculated.

a) Sub-regional weights (used throughout the analysis) correct for gender differences between the survey sample and the population of every country. Besides, the combined weight of responses from a given sub-region corresponds exactly to the share of the relevant youth population in that sub-region relative to the global youth population. As sub-regions, we chose combinations of the five ILO geographical regions and four World Bank income groups (thus, each country is assigned to one of the 20 sub-regions). The weight of individual i who is part of bin b (four bins: 18-29, 30-34, by sex) from sub-region s is calculated as the population in the same bin and sub-region divided by the number of observation in the sample in the same bin and sub-region:

��������_����ℎ ��� =������ ����

���

�1�

b) Country-weights (used for robustness checks, see Section IV.) correct for gender differences between the survey sample and the population of every country. Moreover, the combined weight of responses from a given country corresponds exactly to the share of the relevant youth population in that country relative to the global youth population. The weight of individual i who is part of bin b (four bins: 18-29, 30-34, by sex) from country c is calculated as the population in the same bin and country divided by the number of observation in the sample in the same bin and country:

��� ��_����ℎ ��� =������ ����

���

�2�

c) Within-country weights (used for robustness checks, see section IV.) correct for differences in the share of women/men at the country level but not for differences in response rates across countries. The weight of individual i who is part of bin b (four bins: 18-29, 30-34, by sex) from country c is calculated as the population in the same bin and country divided by the product of the total relevant population in that country (age 18-34) and the number of observation in the sample in the same bin and country:

�� ℎ���� ��_����ℎ ��� =������ ����

������ ��� ∗ ���

�3�

Throughout the analysis, sub-regional weights, equation (1), are used. This is because in the survey sample there are large differences in response rates between countries. The trade-off in the analysis is that including countries with relatively few observations, on the one hand, improves the representativeness at the global level as more countries are included, but on the other hand, introduces increased uncertainty as observations from countries with few participants will receive a relatively large weight.

Sub-regional weights are constructed to address this trade off. Aggregating responses and weights at the sub-regional level reduces considerably the variance of the resulting weights compared to simply using country-level weights (equation (2)).22

IV. Sample representativeness

The section aims to assess the representativeness of the Global Survey on Youth and COVID-19, detailing educational and income profiles of survey respondents. The survey includes observations from 112 countries. In these countries there are altogether 1.47 billion young people (15 to 29 years old) which represent around 92 per cent of the world’s youth population (based on 185 countries for which population data is available).23

Table A-2 presents descriptive statistics on self-reported monthly income before tax in US dollars for a total of 2,111 survey respondents.24 As is common for income data, the distribution is skewed to the right, with an average of USD 1,038/month and a median of USD 444/month.25 Average and median income correlate with the country income group classification. Table A-2 also lists the mean and median GDP per capita (per month) for the same sample. Importantly, GDP per capita and personal gross income cannot be directly compared.26 However it is insightful to observe that the ratio of GDP (column 2) to income (column 1) decreases as the income level of countries rises. This can be seen as an indication that the survey reaches relatively more affluent youth in low- and middle-income countries compared to high-income countries.

22 Sub-regional weights are thus meant to reduce the overall volatility of estimates. Indeed, in the sample at hand the ratio between the observation with lowest weight and the observation with the highest weight is 1:18 (standard deviation: 19.3) while when using weights at country level the ratio is 1:9760 (standard deviation 37.2) 23 We download the data on Working-age population by sex and age (in thousands), from the ILOSTAT explorer. For Jamaica, working-age population is only available for two cohorts 15-24 and 25-34. Then, in this case, we imputed the working-age population for the cohort 15-29 by assuming a uniformly distributed population across age, also within gender. Source: https://www.ilo.org/shinyapps/bulkexplorer47/?lang=en&segment=indicator&id=POP_XWAP_SEX_AGE_NB_A 24 The analysis is based on the question: “What is your monthly income, before tax, in your country's currency?”. In total, 2,111 out of 4,861 (43.4 per cent) of survey respondents aged 20 to 29 who completed the personal information section also provided data on their monthly income. Underreporting might vary across education level, so results should be regarded as indicative only. 25 Huge variations in monthly income across countries and regions suggest that measurement errors might be significant – median values should be more reliable than means. 26 Among other factors, GDP captures all income generated by economic activity in a country, including for example, capital income.

Table A-2: Gross income per month (Global Survey) vs. GDP per capita per month

(1) Gross income

p. month (2) GDP p. capita p.

month

Mean Median Mean Median N

Global 1,038 444 1,175 379 2,111

Country income group

Low income 292 207 66 75 52

Lower-middle income 645 330 225 179 439

Upper-middle income 680 332 596 578 821

High income 2,119 1,407 3,386 3,528 799

Table A-3 compares the educational profile of the survey to that of included countries for the youth cohort aged 20 to 29.27 The first column lists the distribution by education level using ILOSTAT data for 94 of the 112 countries represented in the survey28 which is compared to the self-reported level of education of survey respondents. Survey respondents are considerably more likely to have attained tertiary level education.

Table A-3: Educational profile, youth population (age: 20-29) (%)

(1) Youth population*

(2) Global Survey on Youth and COVID-19 (weighted)

None 12 0

Primary 30 2

Secondary 36 22

Tertiary 21 76

Notes: * 94 countries out of the 112 in the survey sample, data: ILOSTAT. Sub-regional weights apply (see Section II).

Next, table A-4 depicts the share of the youth-to-adult population with advanced education both for the general population (aged 20 – 29) and the survey sample. Overall, 21 per cent of the youth population (aged 20-29), and 24 per cent of those aged 30-34, have attained an advanced education level. Among the (weighted) survey population 76

27 For the purpose of this section we limit the sample to youth aged 20 to 29 years as we can reliably calculate the share for this age range but not for the 18-19 age range (as only data on 15-19 is available). 28 Data on youth working-age population by sex, age, and education (in thousands) are obtained for a total of 143 countries, and matched with the list of 112 countries included in the survey, resulting in 94 countries for which data are available.

per cent (age group: 20-29) to 90 per cent (30-34) have attained tertiary education.29 For the general youth population, a much greater share of young people (aged 20-29) in high-income countries has attained tertiary education (43 per cent) than, for example, in lower middle-income countries (16 per cent). This pattern does not hold for the survey population. In fact, among the survey population young people from lower middle-income countries are most likely to have attained tertiary level education (86 per cent). This is a further indication that in low- and middle-income countries the survey reaches those relatively more educated and privileged compared to in high-income countries.

Overall, these results show that young people (age group: 20-29) in the survey sample are around three times as likely to have attained tertiary education as compared to the general youth population of the same age bracket. In high-income countries, young people with tertiary education are over-represented at a rate of two to one. This is because in these countries the overall level of education among the youth population is higher and because the survey respondents in high-income countries show a more diverse educational profile.

Table A-4: Share of youth-to-adults with advanced education:

General youth population (%)

Global Survey on Youth and COVID-19 (weighted) (%)

20-29*

30-34 20-29 30-34

Global 21 24 76 90

Country income groups

Low-income 5 5 68 87

Lower middle-income 16 15 86 98

Upper middle-income 25 26 77 94

High-income 38 47 80 89

Notes: the age-cohort 20-29 is constructed with 93 out of the 112 included in the survey for which population data is provided by ILOSTAT.

V. Robustness checks

This section presents the results of two different robustness tests. First, we test the sensitivity of the results to changes in the sample – leaving out observations from one country at a time. Second, we compare results when using different weights (as described in Section III). The robustness checks are examining three key outcome variables: the share

29 These descriptive measures were constructed by survey weights (see Section III). This is also why the weighted sample shows a higher share of young people with tertiary education compared to the unweighted sample (see Table A-5). Generally young people from countries that are under-represented in the survey report on average a higher level of education.

of survey respondents who (i) stopped working since the onset of the pandemic; (i) reported that their education will be delayed or might fail and (iii) who were assessed to be probably affected by anxiety or depression.

The first test implies performing the so-called “jackknife procedure”. The method is typically used in statistical regressions but can easily be applied to the framework of the survey. Intuitively, the procedure re-calculates results – here the average of a weighted variable in the sample – leaving observations from one of the 112 countries in the sample out at a time. This tests shows whether the results are primarily driven by one county or a group of countries.30

The second robustness check compares global estimates for the three different weighting approaches described in Section II: (i) the sub-regional weights (used throughout the report), (ii) country-weights and (iii) within-country weights. Country-weights correct for differences in the shares between women and men at the country level and would be the preferred approach in a survey where each country is represented by roughly a similar number of observations (which is not the case in this survey). The within-country weights only correct for within-country gender imbalances but do not correct for different population sizes. Results from these weights can be regarded as baseline estimates (see also table A-6 for an overview on survey participation per country).

The results of the jackknife procedure can be shown in many different ways. Figure A-1 presents them as boxplots, which allows identifying outliers.

Each boxplot shows the same descriptive statistics and can be interpreted similarly:

(1) The average global estimate is represented by a red diamond – in the middle of each boxplot

(2) Each circle represents the re-calculated global estimate leaving aside one country at the time. For better readability, only countries at the maximum and minimum are labelled.

(3) The horizontal lines represent the median (long horizontal line in the middle of each boxplot), the 25th percentile, and the 75th percentile (shorter horizontal lines above and below the median line).

The first column of Panel A in figure A-1 depicts the results of applying the jackknife procedure to the outcome variable “Stopped working” using the preferred weight estimation system (sub-regional weights). There are two clear outliers: China and Sri Lanka. It implies that leaving observations from China aside, the global average for “stopped working” is estimated to be around 18.3 per cent. The opposite is true in the case of Sri Lanka: without the observations from the country, the global average is estimated to be around 15.8 per cent. However, we observe that results when leaving out one of the other 110 countries cluster tightly around the global average of 17.4 per cent (that considers all 112 countries). Panel B and C lead to similar conclusions: leaving one country aside at the time never changes the global estimate by more than 3 percentage points (Panel B), and 1.5 percentage points (Panel C). There are four notable outliers in Panel C – all above the

30 For a more detailed explanation in this procedure see Nisbet R., Miner G., Yale,K. (2018): Chapter 11 - Model Evaluation and Enhancement, Handbook of Statistical Analysis and Data Mining Applications (Second Edition), Academic Press, Pages 215-233.

global estimate suggesting that without those four countries, the global average would be around 5 percentage points higher.

To assess the robustness of our weighting system we compare the global averages (red diamonds) in the three boxplots of each panel. Global estimates vary somewhat by weighting system while not changing results in a qualitatively meaningful manner. As expected outliers in the second column (country-level weights) mostly represent countries with large youth populations while outliers in the third column (within-country weights) represent countries with a large number of respondents in the survey (see also table A-6). Overall, the distribution of country-weights is more spread than for regional-groupings highlighting the sensitivity of this alternative weighting approach.

Overall, survey results appear robust to a variety of sensitivity checks, including different weighting approaches.

Figure A-1: Robustness tests (jacknife procedure and three different weighting methods) for selected outcome variables, (%)

Note: Robustness tests are conducted for the youth sample (age range: 18 to 29) of the survey. Scales in all three panels represent population shares from 0 to 1, corresponding to 0 to 100 per cent.

Table A-5: Sample characteristics: weighted vs. unweighted sample

Weighted sample Unweighted (raw) sample

18-29 30-34 18-29 30-34

Category

No. % No. %

No. % No. %

Labour market status

Working 3,170 28.5 1,027 69.6

2,705 24.3 971 65.8

Studying 5,028 45.2 109 7.4

5,729 51.5 115 7.8

Study and work 1,775 15.9 230 15.6

1,645 14.8 229 15.5

NEET 1157 10.4 109 7.4

1051 9.4 160 10.8

Gender Female 5,958 53.5 788 53.4

7,136 64.1 874 59.3

Male 5,172 46.5 687 46.6

3,994 35.9 601 40.7

Age group

18-24 7,354 66.1

8,071 72.5

25-29 3,776 33.9

3,059 27.5

30-34

1,475 100.0

1,475 100.0

Region

Africa 765 6.9 78 5.3

785 7.1 298 20.2

Americas 2,045 18.4 320 21.7

2,070 18.6 314 21.3

Arab States 141 1.3 10 0.7

178 1.6 29 2.0

Asia and the Pacific 6,018 54.1 734 49.7

2,131 19.1 285 19.3

Europe and Central Asia 2,161 19.4 334 22.6

5,966 53.6 549 37.2

Country income group

Low income 149 1.3 15 1.0

244 2.2 86 5.8

Lower middle income 4,076 36.6 381 25.8

1,966 17.7 412 27.9

Upper middle income 4,142 37.2 620 42.0

4,553 40.9 558 37.8

High income 2,764 24.8 459 31.1

4,367 39.2 419 28.4

Total 11,130 100.0 1,475 100.0

11,130 100.0 1,475 100.0

Marital status

Single 5,783 80.0 405 43.4

5,312 73.5 412 44.1

Married/Have a partner 1068 14.8 493 52.7

1258 17.4 482 51.6

Prefer not to say 379 5.2 37 3.9

660 9.1 41 4.4

Children No 6,797 94.0 632 67.5

6,782 93.8 600 64.2

Yes 433 6.0 303 32.5

448 6.2 335 35.8

Area

Urban 4,277 59.2 614 65.6

4,059 56.1 623 66.6

Suburban 1,574 21.8 217 23.2

1,317 18.2 193 20.6

Rural 1379 19.1 104 11.2

1854 25.6 119 12.7

Highest education level attained

None 64 0.9 8 0.8

131 1.8 8 0.9

Primary 348 4.8 12 1.2

1056 14.6 8 0.9

Secondary 2,057 28.5 78 8.4

2,178 30.1 83 8.9

Tertiary 4,761 65.8 837 89.6

3,865 53.5 836 89.4

Identity

Minority 1461 20.2 172 18.4

1130 15.6 173 18.5

Refugee and migrant 195 2.7 32 3.4

218 3.0 45 4.8

Person with disability 168 2.3 36 3.8

156 2.2 27 2.9

LGBTI 491 6.8 46 4.9

466 6.4 42 4.5

Total 7,230 100.0 935 100.0

7,230 100.0 935 100.0

Table A-6: Observation per country (unweighted sample)

Country 18-29 30-34 Country 18-29 30-34 Country 18-29 30-34 Country 18-29 30-34

Afghanistan 25 5 Croatia 11 1 Kenya 88 22 Romania 81 20

Albania 21 5 Cyprus 7 4 Korea, South 158 9 Russia 59 21

Algeria 15 4 Czech Rep. 12 0 Kyrgyzstan 37 9 Rwanda 10 4

Angola 9 6 DRC 23 7 Latvia 26 3 Saudi Arabia 19 3

Argentina 59 19 Denmark 14 2 Lebanon 73 17 Senegal 32 20

Armenia 38 19 Dom. Rep. 6 4 Malawi 6 5 Sierra Leone 5 11

Australia 31 7 Ecuador 17 6 Malaysia 18 5 Singapore 30 1

Austria 48 19 Egypt 56 9 Maldives 10 3 Slovenia 17 10

Azerbaijan 65 32 El Salvador 47 14 Mali 105 31 South Africa 82 36

Bangladesh 34 31 Fiji 21 8 Mauritania 53 55 Spain 98 31

Barbados 71 13 Finland 18 2 Mexico 307 46 Sri Lanka 248 43

Belgium 88 21 France 197 36 Moldova 25 11 Suriname 24 11

Belize 13 7 Georgia 26 4 Mongolia 21 14 Sweden 20 3

Benin 3 7 Germany 324 45 Morocco 20 2 Switzerland 73 34

Bhutan 12 1 Ghana 15 15 Myanmar 19 11 Tanzania 14 8

Bolivia 47 8 Greece 25 7 Nepal 16 7 Thailand 25 23

Brazil 262 12 Guatemala 200 32 Netherlands 72 5 Trini. & Toba. 101 37

Bulgaria 13 4 Guyana 11 7 New Zealand 9 2 Tunisia 20 13

Burkina Faso 11 19 Honduras 8 4 Nicaragua 5 8 Turkey 140 25

Burundi 9 3 India 559 94 Nigeria 86 63 UK 139 22

Cambodia 33 12 Indonesia 373 64 N. Macedonia 8 2 USA 170 100

Cameroon 7 6 Iraq 60 4 Norway 8 2 Uganda 17 13

Canada 53 7 Ireland 12 4 Pakistan 16 5 Ukraine 82 39

Chile 12 3 Italy 150 34 Panama 55 4 Uruguay 8 3

China 216 22 Jamaica 424 41 Peru 54 12 Venezuela 6 8

Colombia 95 34 Japan 94 41 Philippines 152 34 Vietnam 11 3

Costa Rica 15 15 Jordan 26 25 Poland 2,168 79 Zambia 28 5

Cote d'Ivoire 30 35 Kazakhstan 1,815 222 Portugal 29 7 Zimbabwe 41 9

Contact decentjobsforyouth@ilo.org

Engage www.decentjobsforyouth.org

Follow @decentjobsyouth

Share #decentjobsforyouth

Scaling up action & impact on youth

employment